[FE evolution] - IPv6 HA Integration

[alpersid]

Hello guys,

I am in need of your technical expertise and maybe some advices.
We have the following architecture for our application

As you can see, we have implemented HA for all our elements, but also implemented the autoscaling feature for all elements which brings even a higher value.
We are in big need to integrate the IPv6 in our current architecture, however, is not real easy.

IPv6 was implemented only for ECS as following:

• only for flavor m2 and c2
• only for centos
• with custom image ( redhat) it's not working

The only possible ways I've found to configure HA for IPv6 our current architecture are the followings:

1. Virtual IPv6 for multiple servers

With virtual IPv6 available for multiple servers we can configure keepalived ( or other virtual IP software ) which will switch the IPv6 to another server (MASTER/SLAVE ) in case something will go wrong with the VM.

Limitations:

Virtual IPv6 for multiple servers is not available.

2. IPv6 on ELB

As ELB is a multizone resource, implementing IPv6 on it will solve our need for high availability.

Limitations:

IPv6 on ELB is not available

Our only solution, given the limitation, is to have a single point of failure ( VM ) for IPv6.

As the amount of traffic is really big, and the application is critical, we don't want to configure it until we'll find a more reliable solution

Can you please give me some hints, maybe I've missed something, how can we configure IPv6 in HA mode?

Also, given the current implementation of IPv4, the most suitable will be to be implemented IPv6 in the ELB ( as it's multi AZ and the SLA is 100%).
Do you have any roadmap for the IPv6 implementation for ELB ?

Thank you in advance,
Alex

[ShiChangkuo]

@antonin-a @qukuijin Need your help here, please take a look at this.

[Pseudodevincent]

Hello @alpersid , IPv6 support with ELB will come with Dedicated ELB (ELBv3) feature which is in roadmap for Q4 2021 (implementation has already started).
Until then, maybe you can rely on third party DNS failover mechanisms ?
Sorry, I don't see any other workaround...
Also, IPv6 is supported on m6 and c6 flavors (not m2 and c2) with Centos and Debian.
I'm going to investigate about RedHat problem and come back to you.

[Pseudodevincent]

Hello @alpersid , I just tested OBS RHEL 7.8 public image on a c6 with IPv6 and it seem to work:

[alpersid]

During our tests, after we have rebooted the VMs, we were not able to start them, the status have changed to faulty.
We have also tried to delete them from GUI or API, but it's not working. We have also opened a ticket for the delete issue, but seems that the support team cannot find a way to proceed.

[DafuSHI]

Hello Alexandru @alpersid
My name is Dafu SHI, Cloud Solution Architect for Flexible Engine. I will assist you on this issue. Checking with my collegues I found you have created a support ticket. A fix solution is found and Ops are implementing the fix today. I will keep you updated

[alpersid]

Hello Dafu SHI,

Thanks for your feedback, much appreciated. I confirm that the faulty state issue was solved ( VMs were deleted ).

Regarding the IPv6 implementation, we are now facing another issue with the virtual IP.

Compared with IPv4, the Virtual IPv6 can only be assigned to a single server.

However, the implementation of Virtual IPv6 for a single server seems that is not working properly.

We have assigned multiples virtual IPs to the same NIC, but those IPs are not reachable from internet, only the primary IPv6 is working properly.

Thanks,
Alex

[Pseudodevincent]

Hello @alpersid @DafuSHI , IPv4 VIP are managed by keepalived for HA ECS clusters.
For IPv6 VIP, you are supposed to use VPC API to manage VIP assignement to ECS which requires to code your clustering mechanism...
Regarding the issue with secondary IPv6 VIP not reachable on the internet, can you confirm you assigned a Shared Bandwidth to it ?

[alpersid]

Hello,
@Pseudodevincent - I confirm that for the IPv6 VIP it's working using the APIs, and the issue with the not reachable was related with the Shared Bandwidth. Thank you for your feedback.
So, to conclude, the best HA solution for IPv6 is to wait for ELBv3

[Pseudodevincent]

Hello @alpersid , thanks for the feedback.
I will be testing ELBv3 in preprod this month.
If all ELBv3 tests are OK we will probably be able to release in prod in January.
I'll update you here with the progress.